Merge branch 'gemm_bf16_sk_muozturk' of...

Merge branch 'gemm_bf16_sk_muozturk' of https://github.com/ROCm/composable_kernel into gemm_bf16_sk_muozturk

example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
@@ -152,7 +152,7 @@ int run(int argc, char* argv[])
             break;
         default:
             a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_1<ADataType>{1});
-            b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+            b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Sequential<B0DataType, 1>{});
             b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
         }
 

example/32_batched_gemm_scale_softmax_gemm/run_grouped_query_attention_forward_wmma.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
@@ -156,7 +156,7 @@ int run(int argc, char* argv[])
         b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_2<B1DataType>{-2, 2});
         break;
     default:
-        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<2>{});
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<ADataType, 2>{});
         b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Diagonal<B0DataType>{});
         b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
     }

example/32_batched_gemm_scale_softmax_gemm/run_multi_query_attention_forward_wmma.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
@@ -156,7 +156,7 @@ int run(int argc, char* argv[])
         b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_2<B1DataType>{-2, 2});
         break;
     default:
-        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<2>{});
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<ADataType, 2>{});
         b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Diagonal<B0DataType>{});
         b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
     }

example/32_batched_gemm_scale_softmax_gemm/run_self_attention_wmma.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
@@ -173,7 +173,7 @@ int run(int argc, char* argv[])
         b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_2<B1DataType>{-2, 2});
         break;
     default:
-        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<2>{});
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<ADataType, 2>{});
         b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Diagonal<B0DataType>{});
         b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
     }

example/35_splitK_gemm/run_splitK_gemm_example.inc

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
 #pragma once
 
 struct ProblemSize final
@@ -66,8 +69,8 @@ bool run_splitK_gemm(const ProblemSize& problem_size, const ExecutionConfig& con
         b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
         break;
     default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_Sequential<0>{});
-        b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+        a_m_k.GenerateTensorValue(GeneratorTensor_Sequential<ADataType, 0>{});
+        b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<BDataType, 1>{});
     }
 
     DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());

example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Computes C_m_o = Relu(A0[m, k] * B0[n, k] + D00[m, n] + D01[mn]) * B1[n, o] + D1[m, o]
@@ -60,14 +60,14 @@ struct AddAddRelu
     {
         const ck::half_t x = c + d0 + d1;
 
-        ck::tensor_operation::element_wise::Relu{}.template operator()<ck::half_t>(e, x);
+        ck::tensor_operation::element_wise::Relu{}.operator()(e, x);
     }
     __host__ __device__ void
     operator()(float& e, const float& c, const ck::half_t& d0, const ck::half_t& d1) const
     {
         const float x = c + (d0 + d1);
 
-        ck::tensor_operation::element_wise::Relu{}.template operator()<float>(e, x);
+        ck::tensor_operation::element_wise::Relu{}.operator()(e, x);
     }
 };
 
@@ -377,7 +377,7 @@ int main(int argc, char* argv[])
         break;
     default:
         a0_g_m_k.GenerateTensorValue(GeneratorTensor_1<A0DataType>{1});
-        b0_g_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+        b0_g_k_n.GenerateTensorValue(GeneratorTensor_Sequential<B0DataType, 1>{});
         d00_g_m_n.GenerateTensorValue(GeneratorTensor_1<D00DataType>{1});
         d01_g_m_n.GenerateTensorValue(GeneratorTensor_1<D01DataType>{1});
         b1_g_n_o.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});

example/38_grouped_conv_bwd_data_multiple_d/common.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -41,7 +41,7 @@ struct ExecutionConfig final
 {
     bool do_verification = true;
     int init_method      = 1;
-    bool time_kernel     = true;
+    bool time_kernel     = false;
 };
 
 #define DefaultConvParams                                                                \

example/44_elementwise_permute/CMakeLists.txt

@@ -5,3 +5,4 @@ add_example_executable(example_elementwise_permute_4D_fp32_col elementwise_permu
 add_example_executable(example_elementwise_permute_4D_fp16_col elementwise_permute_4D_fp16_col.cpp)
 add_example_executable(example_elementwise_binary_4D_fp16 elementwise_binary_4D_fp16.cpp)
 add_example_executable(example_elementwise_trinary_4D_fp16 elementwise_trinary_4D_fp16.cpp)
+add_example_executable(elementwise_scale_permute_amax_2D_fp16_fp8 elementwise_scale_permute_amax_2D_fp16_fp8.cpp)

example/44_elementwise_permute/elementwise_scale_permute_amax_2D_fp16_fp8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_dynamic_vector_dims_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+
+#include "ck/library/reference_tensor_operation/cpu/reference_elementwise.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+using F8  = ck::f8_t;
+
+using InputDataType  = F16;
+using ScaleDataType  = F32;
+using OutputDataType = F8;
+
+static constexpr ck::index_t NumDim = 2;
+
+constexpr ck::ReduceTensorOp ReduceOpId = ck::ReduceTensorOp::MAX;
+constexpr bool PropagateNan             = true;
+constexpr bool OutputIndex              = false;
+
+using ReduceOperation = typename ck::reduce_binary_operator<ReduceOpId>::opType;
+
+struct ScalePassThrough
+{
+    ScalePassThrough(const float alpha = 1.f) : alpha_(alpha) {}
+
+    __host__ __device__ constexpr void
+    operator()(OutputDataType& y0, OutputDataType& y1, const InputDataType& x0) const
+    {
+        y0 = ck::type_convert<OutputDataType>(ck::type_convert<ScaleDataType>(x0) * alpha_);
+        y1 = y0;
+    }
+
+    const ScaleDataType alpha_;
+};
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using UnaryAbs    = ck::tensor_operation::element_wise::UnaryAbs;
+
+using DeviceElementwisePermuteInstance = ck::tensor_operation::device::DeviceElementwiseImpl<
+    ck::Tuple<InputDataType>,                  // InDataTypeTuple
+    ck::Tuple<OutputDataType, OutputDataType>, // OutDataTypeTuple
+    ScalePassThrough,                          // Elementwise
+    NumDim,                                    // NumDim
+    256,                                       // BlockSize
+    128,                                       // M0PerBlock
+    128,                                       // M1PerBlock
+    8,                                         // M0PerThread
+    8,                                         // M1PerThread
+    ck::Sequence<1, 0>,                        // ThreadClusterArrangeOrder
+    ck::Sequence<8>,                           // InScalarPerVectorSeq
+    ck::Sequence<8, 1>>;                       // OutScalarPerVectorSeq
+
+using DeviceReduceInstance =
+    ck::tensor_operation::device::DeviceReduceMultiBlock<OutputDataType,
+                                                         ScaleDataType,
+                                                         OutputDataType,
+                                                         NumDim,
+                                                         NumDim,
+                                                         ReduceOperation,
+                                                         UnaryAbs,
+                                                         PassThrough,
+                                                         ck::InMemoryDataOperationEnum::Set,
+                                                         PropagateNan,
+                                                         OutputIndex,
+                                                         false, // HaveIndexInputIfOutputIndex
+                                                         1024,  // BlockSize
+                                                         1,     // MThreadClusterSize
+                                                         1024,  // KThreadClusterSize
+                                                         1,     // MThreadSliceSize
+                                                         16,    // KThreadSliceSize
+                                                         1,     // InSrcVectorDim
+                                                         16,    // InSrceVectorSize
+                                                         1>;    // OutDstVectorSize
+
+void reference_scale_permute_amax(Tensor<InputDataType>& input,
+                                  Tensor<OutputDataType>& host_output_scaled_casted_transposed,
+                                  Tensor<OutputDataType>& host_output_scaled_casted,
+                                  Tensor<OutputDataType>& host_output_amax,
+                                  const float scale)
+{
+    ScalePassThrough out_element_op(scale);
+    const ck::index_t M = input.GetLengths()[0];
+    const ck::index_t K = input.GetLengths()[1];
+    for(ck::index_t m = 0; m < M; m++)
+    {
+        for(ck::index_t k = 0; k < K; k++)
+        {
+            OutputDataType y0, y1;
+            out_element_op(y0, y1, input(m, k));
+
+            host_output_scaled_casted(m, k)            = y0;
+            host_output_scaled_casted_transposed(m, k) = y1;
+            const OutputDataType y_fabs =
+                ck::type_convert<OutputDataType>(ck::math::abs(ck::type_convert<float>(y0)));
+            host_output_amax(0) = ck::type_convert<OutputDataType>(ck::math::max(
+                ck::type_convert<float>(y_fabs), ck::type_convert<float>(host_output_amax(0))));
+        }
+    }
+}
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    bool time_kernel     = true;
+
+    const float scale = 2.f;
+
+    ck::index_t M = 1024;
+    ck::index_t K = 1024;
+
+    if(argc == 3)
+    {
+        M = std::stoi(argv[1]);
+        K = std::stoi(argv[2]);
+    }
+
+    std::array<ck::index_t, 2> dims        = {M, K};
+    std::array<ck::index_t, 2> in_strides  = {K, 1};
+    std::array<ck::index_t, 2> out_strides = {1, M};
+
+    Tensor<InputDataType> input(dims, in_strides);
+    Tensor<OutputDataType> output_scaled_casted_transposed(dims, out_strides);
+    Tensor<OutputDataType> output_scaled_casted(dims, in_strides);
+    Tensor<OutputDataType> output_amax({1});
+
+    input.GenerateTensorValue(GeneratorTensor_3<InputDataType>{0.0, 1.0});
+
+    DeviceMem input_dev_buf(sizeof(InputDataType) * input.mDesc.GetElementSpaceSize());
+    DeviceMem output_scaled_casted_transposed_dev_buf(
+        sizeof(OutputDataType) * output_scaled_casted_transposed.mDesc.GetElementSpaceSize());
+    DeviceMem output_scaled_casted_dev_buf(sizeof(OutputDataType) *
+                                           output_scaled_casted.mDesc.GetElementSpaceSize());
+    DeviceMem output_amax_dev_buf(sizeof(OutputDataType) * output_amax.mDesc.GetElementSpaceSize());
+
+    input_dev_buf.ToDevice(input.mData.data());
+
+    std::array<const void*, 1> inputs = {input_dev_buf.GetDeviceBuffer()};
+    std::array<void*, 2> outputs      = {output_scaled_casted_transposed_dev_buf.GetDeviceBuffer(),
+                                    output_scaled_casted_dev_buf.GetDeviceBuffer()};
+
+    std::cout << "Input: " << input.mDesc << std::endl;
+    std::cout << "Scale: " << scale << std::endl;
+    std::cout << "Output scaled casted transposed: " << output_scaled_casted_transposed.mDesc
+              << std::endl;
+    std::cout << "Output scaled casted: " << output_scaled_casted.mDesc << std::endl;
+    std::cout << "Output amax: " << output_amax.mDesc << std::endl;
+
+    auto launch_transpose_scale = [&]() {
+        auto transposeScale = DeviceElementwisePermuteInstance{};
+        auto argument       = transposeScale.MakeArgumentPointer(dims,
+                                                           {in_strides},
+                                                           {out_strides, in_strides},
+                                                           inputs,
+                                                           outputs,
+                                                           ScalePassThrough{scale});
+
+        if(!transposeScale.IsSupportedArgument(argument.get()))
+        {
+            throw std::runtime_error(
+                "The runtime parameters seems not supported by the device instance, exiting!");
+        };
+
+        auto transposeScale_invoker_ptr = transposeScale.MakeInvokerPointer();
+        return transposeScale_invoker_ptr->Run(argument.get(), StreamConfig{nullptr, time_kernel});
+    };
+
+    auto launch_reduce = [&]() {
+        auto reduce = DeviceReduceInstance{};
+        auto reduce_argument_ptr =
+            reduce.MakeArgumentPointer(dims,
+                                       in_strides,
+                                       {1},    // Output Lengths
+                                       {1},    // Output Strides
+                                       {0, 1}, // Reduce Dims
+                                       static_cast<double>(1.f),
+                                       static_cast<double>(0.f),
+                                       output_scaled_casted_dev_buf.GetDeviceBuffer(),
+                                       nullptr,
+                                       output_amax_dev_buf.GetDeviceBuffer(),
+                                       nullptr,
+                                       UnaryAbs{},
+                                       PassThrough{});
+
+        if(!reduce.IsSupportedArgument(reduce_argument_ptr.get()))
+        {
+            throw std::runtime_error(
+                "The runtime parameters seems not supported by the device instance, exiting!");
+        };
+
+        auto invoker_ptr = reduce.MakeInvokerPointer();
+
+        return invoker_ptr->Run(reduce_argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+    };
+
+    float ave_time = launch_transpose_scale();
+    ave_time += launch_reduce();
+    std::cout << "Perf: " << ave_time << " ms" << std::endl;
+    bool pass = true;
+
+    if(do_verification)
+    {
+        Tensor<OutputDataType> host_output_scaled_casted_transposed(dims, out_strides);
+        Tensor<OutputDataType> host_output_scaled_casted(dims, in_strides);
+        Tensor<OutputDataType> host_output_amax({1});
+
+        reference_scale_permute_amax(input,
+                                     host_output_scaled_casted_transposed,
+                                     host_output_scaled_casted,
+                                     host_output_amax,
+                                     scale);
+
+        output_scaled_casted_transposed_dev_buf.FromDevice(
+            output_scaled_casted_transposed.mData.data());
+        output_scaled_casted_dev_buf.FromDevice(output_scaled_casted.mData.data());
+        output_amax_dev_buf.FromDevice(output_amax.mData.data());
+
+        pass &= ck::utils::check_err(output_scaled_casted_transposed.mData,
+                                     host_output_scaled_casted_transposed.mData,
+                                     "Error: Incorrect results scaled transposed",
+                                     1e-3,
+                                     1e-3);
+        pass &= ck::utils::check_err(output_scaled_casted.mData,
+                                     host_output_scaled_casted.mData,
+                                     "Error: Incorrect results scaled",
+                                     1e-3,
+                                     1e-3);
+        pass &= ck::utils::check_err(
+            output_amax.mData, host_output_amax.mData, "Error: Incorrect results amax", 1e-3, 1e-3);
+    }
+
+    return pass ? 0 : 1;
+}

example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute_xdl.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
@@ -248,7 +248,7 @@ int main(int argc, char* argv[])
         d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_1<D0DataType>{1});
         break;
     default:
-        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<2>{});
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<ADataType, 2>{});
         b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Diagonal<B0DataType>{});
         b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
         d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_1<D0DataType>{1});

example/59_grouped_gemm_multi_ABD/grouped_gemm_multi_abd_xdl_fixed_nk_bias_bf16_i8.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -194,9 +194,9 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
             b1_tensors[i].GenerateTensorValue(GeneratorTensor_3<B1DataType>{-0.5, 0.5});
             break;
         default:
-            a0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{});
-            b0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{});
-            b1_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+            a0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<A0DataType, 0>{});
+            b0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<B0DataType, 1>{});
+            b1_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<B1DataType, 1>{});
         }
 
         d0_tensors[i].GenerateTensorValue(GeneratorTensor_3<D0DataType>{-0.5, 0.5});

example/59_grouped_gemm_multi_ABD/grouped_gemm_multi_abd_xdl_fixed_nk_bias_fp16.cpp

@@ -184,9 +184,9 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
             b_tensors[i].GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
             break;
         default:
-            a0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{});
-            a1_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{});
-            b_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+            a0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<A0DataType, 0>{});
+            a1_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<A1DataType, 0>{});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<B0DataType, 1>{});
         }
 
         d0_tensors[i].GenerateTensorValue(GeneratorTensor_3<D0DataType>{-0.5, 0.5});

example/62_convnd_activ/CMakeLists.txt

@@ -6,6 +6,7 @@ add_subdirectory(convscale_add)
 add_subdirectory(convscale_reduce)
 add_subdirectory(multi_AB)
 add_subdirectory(unary)
+add_subdirectory(dynamic_unary)
 
 add_custom_target(example_convnd_activ_xdl)
 # ScaleAdd ScaleAdd Relu

example/62_convnd_activ/convscale/convnd_fwd_convscale_common.hpp

@@ -172,12 +172,13 @@ bool run_grouped_conv_fwd(bool do_verification,
     {
     case 0: break;
     case 1:
-        in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
-        wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
+        // values generated: -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5
+        in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 6});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-1.0, 1.0});
         break;
     default:
-        in.GenerateTensorValue(GeneratorTensor_3<InDataType>{0.0, 1.0});
-        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
+        in.GenerateTensorValue(GeneratorTensor_3<InDataType>{-5.0, 5.0});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-1.0, 1.0});
     }
 
     DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());

example/62_convnd_activ/dynamic_unary/CMakeLists.txt

+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+      add_custom_target(example_convnd_activ_dynamic_unary_xdl)
+      # Sigmoid
+      add_example_executable(example_convnd_fwd_xdl_dynamic_sigmoid_fp16 convnd_fwd_xdl_dynamic_sigmoid_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_sigmoid_fp16)
+      # Tanh
+      add_example_executable(example_convnd_fwd_xdl_dynamic_tanh_fp16 convnd_fwd_xdl_dynamic_tanh_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_tanh_fp16)
+      # Relu
+      add_example_executable(example_convnd_fwd_xdl_dynamic_relu_fp16 convnd_fwd_xdl_dynamic_relu_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_relu_fp16)
+      # SoftRelu
+      add_example_executable(example_convnd_fwd_xdl_dynamic_softrelu_fp16 convnd_fwd_xdl_dynamic_softrelu_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_softrelu_fp16)
+      # Abs
+      add_example_executable(example_convnd_fwd_xdl_dynamic_abs_fp16 convnd_fwd_xdl_dynamic_abs_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_abs_fp16)
+      # Pow
+      add_example_executable(example_convnd_fwd_xdl_dynamic_pow_fp16 convnd_fwd_xdl_dynamic_pow_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_pow_fp16)
+      # Clipped Relu
+      add_example_executable(example_convnd_fwd_xdl_dynamic_clippedrelu_fp16 convnd_fwd_xdl_dynamic_clippedrelu_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_clippedrelu_fp16)
+      # Leaky Relu
+      add_example_executable(example_convnd_fwd_xdl_dynamic_leakyrelu_fp16 convnd_fwd_xdl_dynamic_leakyrelu_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_leakyrelu_fp16)
+      # Elu
+      add_example_executable(example_convnd_fwd_xdl_dynamic_elu_fp16 convnd_fwd_xdl_dynamic_elu_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_elu_fp16)
+      # Swish
+      add_example_executable(example_convnd_fwd_xdl_dynamic_swish_fp16 convnd_fwd_xdl_dynamic_swish_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_swish_fp16)
+      # PassThrough
+      add_example_executable(example_convnd_fwd_xdl_dynamic_passthrough_fp16 convnd_fwd_xdl_dynamic_passthrough_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_passthrough_fp16)
+      # Logistic
+      add_example_executable(example_convnd_fwd_xdl_dynamic_logistic_fp16 convnd_fwd_xdl_dynamic_logistic_fp16.cpp)
+      add_example_dependencies(example_convnd_activ_dynamic_unary_xdl example_convnd_fwd_xdl_dynamic_logistic_fp16)
+   
+   set(target 1)
+ endif()
+endforeach()

example/62_convnd_activ/dynamic_unary/convnd_fwd_activ_dynamic_unary_common.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <iostream>
+#include <numeric>
+#include <type_traits>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+constexpr ck::index_t NDimSpatial = 3;
+using InDataType                  = ck::half_t;
+using WeiDataType                 = ck::half_t;
+using AccDataType                 = float;
+using CShuffleDataType            = ck::half_t;
+using OutDataType                 = ck::half_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using InLayout  = ck::tensor_layout::convolution::GNDHWC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout = ck::tensor_layout::convolution::GNDHWK;
+
+using InElementOp      = ck::tensor_operation::element_wise::PassThrough;
+using WeiElementOp     = ck::tensor_operation::element_wise::PassThrough;
+using DynamicElementOp = ck::tensor_operation::element_wise::DynamicUnaryOp;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceGroupedConvNDActivInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        DynamicElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        32,          // KPerBlock
+        8,           // AK1
+        8,           // BK1
+        32,          // MPerXdl
+        32,          // NPerXdl
+        2,           // MXdlPerWave
+        4,           // NXdlPerWave
+        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
+        2,           // ABlockTransferSrcVectorDim
+        8,           // ABlockTransferSrcScalarPerVector
+        8,           // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        8,           // BBlockTransferSrcScalarPerVector
+        8,           // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 32, 1, 8>,
+        8>;
+
+template <ck::index_t NDimSpatial,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename InElementOp,
+          typename WeiElementOp,
+          typename OutElementOp,
+          typename DeviceConvNDFwdInstance>
+bool run_grouped_conv(bool do_verification,
+                      int init_method,
+                      bool time_kernel,
+                      const ck::utils::conv::ConvParam& conv_param,
+                      const HostTensorDescriptor& in_g_n_c_wis_desc,
+                      const HostTensorDescriptor& wei_g_k_c_xs_desc,
+                      const HostTensorDescriptor& out_g_n_k_wos_desc,
+                      const InElementOp& in_element_op,
+                      const WeiElementOp& wei_element_op,
+                      const OutElementOp& out_element_op)
+{
+
+    Tensor<InDataType> in(in_g_n_c_wis_desc);
+    Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
+    Tensor<OutDataType> out_host(out_g_n_k_wos_desc);
+    Tensor<OutDataType> out_device(out_g_n_k_wos_desc);
+
+    std::cout << "in: " << in.mDesc << std::endl;
+    std::cout << "wei: " << wei.mDesc << std::endl;
+    std::cout << "out: " << out_host.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-2, 2});
+        wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-2, 2});
+        break;
+    default:
+        in.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.05, 0.05});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
+    DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in.mData.data());
+    wei_device_buf.ToDevice(wei.mData.data());
+
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+    copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
+    copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);
+    copy(wei_g_k_c_xs_desc.GetLengths(), b_g_k_c_xs_lengths);
+    copy(wei_g_k_c_xs_desc.GetStrides(), b_g_k_c_xs_strides);
+    copy(out_g_n_k_wos_desc.GetLengths(), e_g_n_k_wos_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), e_g_n_k_wos_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    // do Conv
+    auto conv     = DeviceConvNDFwdInstance{};
+    auto invoker  = conv.MakeInvoker();
+    auto argument = conv.MakeArgument(in_device_buf.GetDeviceBuffer(),
+                                      wei_device_buf.GetDeviceBuffer(),
+                                      std::array<const void*, 0>{},
+                                      out_device_buf.GetDeviceBuffer(),
+                                      a_g_n_c_wis_lengths,
+                                      a_g_n_c_wis_strides,
+                                      b_g_k_c_xs_lengths,
+                                      b_g_k_c_xs_strides,
+                                      std::array<std::array<ck::index_t, NDimSpatial + 3>, 0>{{}},
+                                      std::array<std::array<ck::index_t, NDimSpatial + 3>, 0>{{}},
+                                      e_g_n_k_wos_lengths,
+                                      e_g_n_k_wos_strides,
+                                      conv_filter_strides,
+                                      conv_filter_dilations,
+                                      input_left_pads,
+                                      input_right_pads,
+                                      in_element_op,
+                                      wei_element_op,
+                                      out_element_op);
+
+    if(!conv.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error("The device op with the specified compilation parameters does "
+                                 "not support this convolution problem.");
+    }
+
+    float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop      = conv_param.GetFlops();
+    std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
+
+    float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+    float gb_per_sec = num_btype / 1.E6 / avg_time;
+    std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << conv.GetTypeString() << std::endl;
+
+    if(do_verification)
+    {
+        auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
+                                                                     InDataType,
+                                                                     WeiDataType,
+                                                                     OutDataType,
+                                                                     InElementOp,
+                                                                     WeiElementOp,
+                                                                     OutElementOp>();
+
+        auto ref_invoker  = ref_conv.MakeInvoker();
+        auto ref_argument = ref_conv.MakeArgument(in,
+                                                  wei,
+                                                  out_host,
+                                                  conv_param.conv_filter_strides_,
+                                                  conv_param.conv_filter_dilations_,
+                                                  conv_param.input_left_pads_,
+                                                  conv_param.input_right_pads_,
+                                                  in_element_op,
+                                                  wei_element_op,
+                                                  out_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        out_device_buf.FromDevice(out_device.mData.data());
+
+        return ck::utils::check_err(out_device, out_host, "Error: incorrect results!");
+    }
+
+    return true;
+}

example/62_convnd_activ/dynamic_unary/convnd_fwd_xdl_dynamic_abs_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_dynamic_unary_common.hpp"
+
+#include "../run_convnd_activ_dynamic_example.inc"
+
+int main(int argc, char* argv[])
+{
+
+    ck::tensor_operation::element_wise::UnaryAbs out_element_op;
+    return !run_convnd_example(argc, argv, out_element_op);
+}

example/62_convnd_activ/dynamic_unary/convnd_fwd_xdl_dynamic_clippedrelu_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_dynamic_unary_common.hpp"
+
+#include "../run_convnd_activ_dynamic_example.inc"
+
+int main(int argc, char* argv[])
+{
+
+    ck::tensor_operation::element_wise::ClippedRelu out_element_op(0.f, 1.f);
+    return !run_convnd_example(argc, argv, out_element_op);
+}

example/62_convnd_activ/dynamic_unary/convnd_fwd_xdl_dynamic_elu_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_dynamic_unary_common.hpp"
+
+#include "../run_convnd_activ_dynamic_example.inc"
+
+int main(int argc, char* argv[])
+{
+
+    ck::tensor_operation::element_wise::Elu out_element_op(2.f);
+    return !run_convnd_example(argc, argv, out_element_op);
+}

example/62_convnd_activ/dynamic_unary/convnd_fwd_xdl_dynamic_leakyrelu_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_dynamic_unary_common.hpp"
+
+#include "../run_convnd_activ_dynamic_example.inc"
+
+int main(int argc, char* argv[])
+{
+
+    ck::tensor_operation::element_wise::LeakyRelu out_element_op(0.f);
+    return !run_convnd_example(argc, argv, out_element_op);
+}